The aesthetic quality of a scene depends strongly on camera viewpoint. Existing approaches for aesthetic viewpoint suggestion are either single-view adjustments, predicting limited camera adjustments from a single image without understanding scene geometry, or 3D exploration approaches, which rely on dense captures or prebuilt 3D environments coupled with costly reinforcement learning (RL) searches. In this work, we introduce the notion of 3D aesthetic field that enables geometry-grounded aesthetic reasoning in 3D with sparse captures, allowing efficient viewpoint suggestions in contrast to costly RL searches. We opt to learn this 3D aesthetic field using a feedforward 3D Gaussian Splatting network that distills high-level aesthetic knowledge from a pretrained 2D aesthetic model into 3D space, enabling aesthetic prediction for novel viewpoints from only sparse input views. Building on this field, we propose a two-stage search pipeline that combines coarse viewpoint sampling with gradient-based refinement, efficiently identifying aesthetically appealing viewpoints without dense captures or RL exploration. Extensive experiments show that our method consistently suggests viewpoints with superior framing and composition compared to existing approaches, establishing a new direction toward 3D-aware aesthetic modeling.
Figures & tables
Figure 2: We distill aesthetic features into a feedforward Gaussian Splatting network ( top ). At inference, to search for aesthetic viewpoints, we adopt a two-stage pipeline: coarse sampling to find good candidates ( bottom left ) and local refinement by gradient ascent ( bottom right ).
Table 1: Novel view aesthetic score correlation between predicted and ground-truth scores.
Figure 3: (a) Aesthetic score predictions over consecutive frames. Our method produces scores closer to the ground truth while being smoother and more consistent across nearby views. The dashed line and box mark the regions visualized in (b) and (c), respectively. (b) Given the same aesthetic model and viewpoint, rendering artifacts in the predicted view ( bottom ) bias the RGB-scoring approach toward lower scores. (c) Ground truth scores fluctuate noticeably across nearly identical nearby views.
RE10k [ 41 ]
Methods
2 Input Views
4 Input Views
6 Input Views
VEN ↑
SAMPNet ↑
VEN ↑
SAMPNet ↑
VEN ↑
SAMPNet ↑
Baseline
1.48
2.29
1.79
2.26
2.01
2.29
In-plane Shift ∗
1.52
2.31
1.81
2.36
2.10
2.41
Rotation ∗
1.78
2.38
1.95
2.42
2.13
2.45
UNIC [ 14 ] †
1.15
2.17
1.61
2.33
1.82
2.35
Table 2: Aesthetic viewpoint suggestion evaluation on different datasets and with different numbers of input views. † indicates open-sourced methods adapted to our setting. ∗ denotes approximations of non–open-sourced single-view methods in our setting. Note that such approximations may overestimate the actual results that those methods could deliver.
Figure 4: Aesthetic viewpoint suggestion examples with 4 input views in RE10k ( top 2 rows ) and DL3DV ( bottom 3 rows ). Red boxes in the last row show that single-view methods fail to remove distracting objects due to limited adjustment range. Zoom in for more details.
Figure 5: Visualization of sampled viewpoints colored by aesthetic score, with representative renderings shown alongside.
Methods
RE10k [ 41 ]
DL3DV [ 13 ]
Δ VEN ↑
Δ VEN ↑
Baseline
0.20
0.18
Ours
0.46
0.43
Table 3: Aesthetic score improvement with gradient ascent.
Figure 6: Gradient ascent results. Zoom in for a closer look.
Methods
RE10k [ 41 ]
DL3DV [ 13 ]
PLCC
SRCC
PLCC
SRCC
w/o. view-cond.
0.732
0.695
0.658
0.625
w. view-cond.
0.796
0.758
0.700
0.668
Table 4: Ablation of view-conditioning on novel view aesthetic prediction with 4 input views.
# Candidates
RE10k [ 41 ]
DL3DV [ 13 ]
VEN ↑
VEN ↑
1
1.96
2.57
2
2.03
2.76
3
2.05
2.78
# Steps
Δ VEN ↑
Δ VEN ↑
15
0.21
0.15
Table 5: Ablation of candidate numbers and refinement steps.
Figure S1 : Additional novel view aesthetic prediction results. In both examples: (a) Aesthetic score predictions over consecutive frames. The dashed line and box mark the regions visualized in (b) and (c), respectively. (b) Given the same aesthetic model and viewpoint, rendering artifacts in the predicted view ( e.g . , wrong color prediction as marked in red boxes in top , noisy predictions marked in blue boxes in bottom ) bias the RGB-scoring approach. (c) Ground-truth scores fluctuate noticeably across nearly identical nearby views.
Figure S2 : Additional results of aesthetic viewpoint suggestion.
Figure S3 : Additional visualizations of sampled viewpoints colored by aesthetic score, with representative renderings shown alongside.
Figure S4 : Additional gradient ascent results. Zoom in for a closer look.
S=4
S=8
S=16
S=32
N=4
1.98
2.00
1.98
1.98
N=8
1.98
1.94
2.03
2.02
N=16
2.01
2.00
2.02
2.04
N=32
2.02
2.03
2.04
2.06
Table S1: Ablation of the sampling configuration of search stage 1 on novel view aesthetic prediction. S denotes the number of samples along each segment of the interpolated input-view trajectory, and N denotes the number of neighboring viewpoints sampled around each segment-sample. All experiments use 4 input views on RE10k [ 41 ] , and the aesthetic quality is measured by VEN [ 29 ] .
As 3D Gaussian Splatting (3DGS) gains attention in immersive media and digital content creation, assessing the aesthetics of 3D scenes becomes important in helping creators build more visually compelling 3D content. However, existing evaluation methods for 3D scenes primarily emphasize reconstruction fidelity and perceptual realism, largely overlooking higher-level aesthetic attributes such as composition, harmony, and visual appeal. This limitation comes from two key challenges: (1) the absence of general 3DGS datasets with aesthetic annotations, and (2) the intrinsic nature of 3DGS as a low-level primitive representation, which makes it difficult to capture high-level aesthetic features. To address these challenges, we propose Aes3D, the first systematic framework for assessing the aesthetics of 3D neural rendering scenes. Aes3D includes Aesthetic3D, the first dataset dedicated to 3D scene aesthetic assessment, built on our proposed annotation strategy for 3D scene aesthetics. In addition, we present Aes3DGSNet, a lightweight model that directly predicts scene-level aesthetic scores from 3DGS representations. Notably, our model operates solely on 3D Gaussian primitives, eliminating the need for rendering multi-view images and thus reducing computational cost and hardware requirements. Through aesthetics-supervised learning on multi-view 3DGS scene representations, Aes3DGSNet effectively captures high-level aesthetic cues and accurately regresses aesthetic scores. Experimental results demonstrate that our approach achieves strong performance while maintaining a lightweight design, establishing a new benchmark for 3D scene aesthetic assessment. Code and datasets will be made available in a future version.
Chuanzhi Xu, Boyu Wei, Haoxian Zhou +5
The University of Sydney · The University of Hong Kong
Generating explorable 3D scenes from a single image is a highly challenging problem in 3D vision. Existing methods struggle to support free exploration, often producing severe geometric distortions and noisy artifacts when the viewpoint moves far from the original perspective. We introduce One2Scene, an effective framework that decomposes this ill-posed problem into three tractable sub-tasks to enable immersive explorable scene generation. We first use a panorama generator to produce anchor views from a single input image as initialization. Then, we lift these 2D anchors into an explicit 3D geometric scaffold via a generalizable, feed-forward Gaussian Splatting network. Instead of treating the panorama as a single image for reconstruction, we project it into multiple sparse anchor views and reformulate the reconstruction task as multi-view stereo matching, which allows us to leverage robust geometric priors learned from large-scale multi-view datasets. A bidirectional feature fusion module is used to enforce cross-view consistency, yielding an efficient and geometrically reliable scaffold. Finally, the scaffold serves as a strong prior for a novel view generator to produce photorealistic and geometrically accurate views at arbitrary cameras. By explicitly conditioning on a 3D-consistent scaffold to perform reconstruction, One2Scene works stably under large camera motions, supporting immersive scene exploration. Extensive experiments show that One2Scene substantially outperforms state-of-the-art methods in panorama depth estimation, feed-forward 360° reconstruction, and explorable 3D scene generation. Project page: https://one2scene5406.github.io/
We introduce ReconSplat, a feed-forward model for 3D scene reconstruction that aims to address the longstanding trade-off between plausible view generation for unobserved regions and geometric consistency, providing both geometrically aligned novel views and sharp depth estimates. Our approach builds on 3D Gaussian splatting (3DGS) as an intermediate differentiable scene representation and integrates it with a multi-view latent diffusion model (MV-LDM) trained to act simultaneously as a refiner and an inpainter for appearance and scene geometry. We enforce geometric consistency by guiding the diffusion process with variational 3D latent features for appearance and geometry, encoded by the feed-forward 3DGS representation and rasterized to 2D latent space. ReconSplat produces both photorealistic novel views and accurate depth maps on real-world benchmarks, RealEstate10K and DL3DV-10K, outperforming existing methods in challenging extrapolation setups. Notably, ReconSplat allows the extrapolation of unseen and challenging viewpoints jointly with coherent and precise scene geometry.
Giuseppe Stracquadanio, Kevin Raj, Julia Grabinski +1